ubic.basecode.math

Class MultipleTestCorrection

java.lang.Object

ubic.basecode.math.MultipleTestCorrection

public class MultipleTestCorrection
extends Object

Methods for p-value correction of sets of hypothesis tests.

FIXME make this API more consistent.

Author:

Paul Pavlidis

Constructor Summary

Constructors

Constructor and Description
MultipleTestCorrection()

Method Summary

Modifier and Type	Method and Description
static cern.colt.list.DoubleArrayList	benjaminiHochberg(cern.colt.list.DoubleArrayList pvalues)
static cern.colt.matrix.DoubleMatrix1D	benjaminiHochberg(cern.colt.matrix.DoubleMatrix1D pvalues)
static double	benjaminiHochbergCut(cern.colt.list.DoubleArrayList pvalues, double fdr) Benjamini-Hochberg method.
static double	BenjaminiYekuteliCut(cern.colt.list.DoubleArrayList pvalues, double fdr) Benjamini-Yekuteli method.
static double	BonferroniCut(cern.colt.list.DoubleArrayList pvalues, double fwe) Determine the Bonferroni pvalue threshold to maintain the family wise error rate (assuming pvalues are independent).

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

MultipleTestCorrection

public MultipleTestCorrection()

Method Detail

benjaminiHochberg

public static cern.colt.list.DoubleArrayList benjaminiHochberg(cern.colt.list.DoubleArrayList pvalues)

Parameters:

pvalues -

Returns:

false discovery rates (FDR) computed using the method of Benjamini and Hochberg. The order is the same as the original pvalues (that is, each value fdr[i] corresponds to the pvaluesp[i].

benjaminiHochberg

public static cern.colt.matrix.DoubleMatrix1D benjaminiHochberg(cern.colt.matrix.DoubleMatrix1D pvalues)

Parameters:

pvalues; - can contain missing values or invalid pvalues (outside range [0-1]), which are ignored.

Returns:

false discovery rates (FDRs) computed using the method of Benjamini and Hochberg; or null if they could not be computed. The order is the same as the original pvalues (that is, each value fdr[i] corresponds to the pvaluesp[i].

benjaminiHochbergCut

public static double benjaminiHochbergCut(cern.colt.list.DoubleArrayList pvalues,
                                          double fdr)

Benjamini-Hochberg method. Determines the maximum p value to maintain the false discovery rate. (Assuming pvalues are independent);

Parameters:

pvalues - list of pvalues. Need not be sorted.

fdr - false discovery rate (value q in B-H).

Returns:

The maximum pvalue that maintains the false discovery rate

Throws:

IllegalArgumentException - if invalid pvalues are encountered.

BenjaminiYekuteliCut

public static double BenjaminiYekuteliCut(cern.colt.list.DoubleArrayList pvalues,
                                          double fdr)

Benjamini-Yekuteli method. Determines the maximum p value to maintain the false discovery rate. Valid under dependence of the pvalues. This method is more conservative than Benjamini-Hochberg.

Parameters:

pvalues - list of pvalues. Need not be sorted.

fdr - false discovery rate (value q in B-H).

Returns:

The maximum pvalue that maintains the false discovery rate

Throws:

IllegalArgumentException - if invalid pvalues are encountered.

BonferroniCut

public static double BonferroniCut(cern.colt.list.DoubleArrayList pvalues,
                                   double fwe)

Determine the Bonferroni pvalue threshold to maintain the family wise error rate (assuming pvalues are independent).

Parameters:

pvalues - The pvalues

fwe - The family wise error rate

Returns:

The minimum pvalue that maintains the FWE